Khanmigo Results: Why Sal Khan Called It 'a Non-Event'

Three years ago, Sal Khan told a TED audience that AI would bring “probably the biggest positive transformation that education has ever seen,” with every student getting “an artificially intelligent but amazing personal tutor.” In April 2026 he gave a different summary of the Khanmigo results to Chalkbeat: “For a lot of students, it was a non-event. They just didn’t use it much.” It is rare for a founder to say that about his own product, and it deserves attention. The problem Khan describes isn’t that the AI was bad at tutoring. It’s that most kids never asked it to tutor them.

Key Takeaways

• Sal Khan told Chalkbeat (April 9, 2026) that Khanmigo was “a non-event” for many students: “Some will; most won’t” seek out help.
• Khan Academy’s chief learning officer said the core issue is that “students aren’t great at asking questions well,” and AI can only respond to what it’s asked.
• A teacher at an early-adopting Indiana high school found students frustrated that Khanmigo sometimes made mistakes but wouldn’t give away answers; she stopped using it.
• Khan Academy has since built Khanmigo directly into practice problems because “students were not seeking out Khanmigo’s help as much as we had hoped.”
• Stanford’s 2026 evidence review found no high-quality causal studies of AI tools for US K–12 students; for parents, the takeaway is that AI tutors need a structure and a human around them.

What Sal Khan actually said

Khanmigo is Khan Academy’s AI tutor and teaching assistant, built on OpenAI’s GPT-4 and launched in 2023. It was designed to guide students rather than give answers. Khan was its most visible champion, and his 2024 book was titled Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing).

In an interview with Chalkbeat’s ideas editor Matt Barnum, published April 9, 2026, Khan reflected on what happened in classrooms.

“For a lot of students, it was a non-event,” he said. “They just didn’t use it much.”

He offered an analogy. Imagine he walked into a class, sat in the back of the room and waited for students to come ask for help. “Some will; most won’t.” That, he said, has been the experience with AI tutoring. It doesn’t necessarily make students motivated to learn, or fill in the gaps in knowledge they’d need to ask good questions.

He hasn’t given up on AI. “I just view it as part of the solution; I don’t view it as the end-all and be-all,” he said. And on the company’s redesign: “AI is going to help. But I think our biggest lever is really investing in the human systems.”

Kristen DiCerbo, Khan Academy’s chief learning officer, was blunter: “So far I am not seeing the revolution in education.”

Why an AI tutor that waits to be asked mostly goes unused

The engineering reason is simple once you see it. A chatbot tutor is a pull system. Nothing happens until the student types a question. That design quietly assumes three things about the student:

1. They notice they’re stuck.
2. They want help, rather than wanting to be done.
3. They know enough to ask a useful question.

Each assumption fails for many kids. DiCerbo put the third one plainly: AI can only respond to students based on what they ask, and “students aren’t great at asking questions well.” A student who doesn’t understand a math concept often can’t describe what they don’t understand. That’s why they’re stuck.

The second assumption collides with a design choice that made Khanmigo responsible. To prevent cheating, it was restricted from simply giving answers. Kristen Musall, a geometry teacher at Hobart High School in Indiana, an early adopter featured on 60 Minutes, told Chalkbeat that students found it frustrating: Khanmigo “sometimes made mistakes, but also wouldn’t give away the answer.” Her conclusion: “If students don’t engage with the material enough to know what they’re looking for, then an AI like Khanmigo doesn’t necessarily help.” She no longer uses it, and she says administrators have been more enthusiastic about it than teachers.

There’s a hard trade-off here. An AI that gives answers gets used but can hurt learning; a 2025 randomized trial in PNAS found high schoolers who used a plain GPT-4 chatbot on practice problems later scored 17% lower on an exam without it. An AI that withholds answers protects learning but frustrates kids who just want to be finished. Neither design, on its own, solves motivation.

Khanmigo results in context: the same pattern in other studies

Khan’s experience isn’t an outlier. In a June 2026 Stanford study, elementary students given dedicated time with an AI reading tutor averaged only 2 to 5 minutes of use a week out of a planned 60, and about half never logged on. Students who did use it on their own tended to be higher-achieving. We cover that research in our article on AI reading tutors kids barely used.

The broader evidence is thin. Stanford SCALE’s 2026 review of AI in K–12 found that of more than 800 relevant papers, only 20 produced strong causal evidence, and the authors “did not identify any high-quality causal studies in K-12 settings in the U.S. for students.” Chalkbeat cited the same report’s conclusion that the evidence base remains “extremely limited.” Where AI did help in the studies that exist, “tools designed with pedagogical guardrails… show more promise than general purpose AI tools.”

Here’s how the claims and the evidence compare:

Claim about AI tutors	What the evidence says now	Source
Every student will have a personal AI tutor	Many students given one rarely use it	Khan to Chalkbeat, 2026; Stanford reading tutor trials, 2026
AI will personalize learning to each child	Not yet observed by Khan Academy’s own learning chief	DiCerbo to Chalkbeat, 2026
Guarded AI tutors protect learning	Supported in one large high school trial; no exam gains	Bastani et al., PNAS, 2025
Well-designed AI tutors can beat good classes	Supported in one Harvard college physics RCT	Kestin et al., Scientific Reports, 2025
Rigorous K–12 evidence on AI exists	20 strong causal papers out of 800+; none for US K–12 students	Stanford SCALE review, 2026
Practice platforms (pre-AI) help	Classes using Khan Academy made slightly faster math gains	Eames et al., PNAS, 2026

That last row is worth noticing. A 2026 study in PNAS by Taryn Eames, Emma Brunskill, Philip Oreopoulos and colleagues, using three years of administrative data, found that “even marginal” use of Khan Academy’s practice platform improved student math performance, with students who had higher prior achievement benefiting more. That research predates Khanmigo. The structured practice worked; the chat-on-demand layer is what students skipped.

What Khan Academy changed

Khan Academy’s response is the most useful part of the story. Instead of waiting for students to open a chat window, the company built Khanmigo into the practice itself. A spokesperson told Chalkbeat the change was made because “students were not seeking out Khanmigo’s help as much as we had hoped.”

In its announcement of the redesigned platform, Khan Academy described Khanmigo as “a learning coach” that “provides one-to-one student support when learners get stuck,” embedded directly into the platform “not tacked on,” with the reimagined experience available to district partners for back-to-school 2026. The company was candid about its own image: one of its leaders compared Khan Academy to Brussels sprouts, good for kids but not always appealing.

In engineering terms, they moved from pull to push. The help appears at the moment a student gets a problem wrong, rather than waiting to be asked. Whether that design produces better learning hasn’t been independently tested yet. It’s a reasonable bet, and it’s a company’s description of its own product.

What parents should take from this

Don’t assume access means use

If your child’s school says students “have access” to an AI tutor, ask how often they use it. Platforms track this. A tool that exists in the school’s software list but sees two minutes a week isn’t helping anyone.

Teach the skill of asking a good question

This is the hidden skill behind every tutor, human or AI. Practice it at home. When your child is stuck, before helping, ask: “What exactly don’t you get? Which step?” A child who can say “I don’t know why you flip the second fraction” can use any tutor well. A child who can only say “I don’t get it” can’t.

Put structure around the tool

Khan Academy’s own fix was structure: help that arrives inside the practice. At home, you can add structure too. Set a short, regular time. Sit nearby. Ask afterward what the tutor helped with. Khan’s point about “human systems” applies to families as much as schools.

Expect frustration with tutors that won’t give answers

If your child complains that the AI “won’t just tell me,” that’s the guardrail doing its job. Acknowledge the frustration, then help them form a better question. Our comparison of SchoolAI, Khanmigo and similar classroom tools explains how these guardrails differ.

What not to do: treat an AI tutor as a substitute for help your child needs

The students most likely to skip AI tutors are the ones who struggle most. If your child is behind, a human tutor, teacher or intervention specialist has far stronger evidence. Large public investments in AI tutoring, like the one examined in our article on North Carolina’s Khanmigo contract, are worth asking hard questions about.

What to Watch For Over the Next 3 Months

• Week 4: Ask your child whether they’ve used the school’s AI tutor this month and for what. If they don’t know what you mean, that’s your answer.
• Month 2 red flags: Your child saying the AI is “useless” because it won’t give answers, switching to a general chatbot that will, or a struggling student being pointed to the AI instead of to a person.
• Month 3 self-check: Can your child describe exactly where they get stuck in a subject? Does the school know how much its AI tutor is used? Has anyone measured whether it helps?

Frequently Asked Questions

What did Sal Khan say about Khanmigo?

In an April 2026 Chalkbeat interview, Khan said that “for a lot of students, it was a non-event. They just didn’t use it much.” He compared it to a tutor sitting at the back of a classroom: “Some will; most won’t” ask for help. He still sees AI as “part of the solution,” not the whole answer.

Does Khanmigo work?

There’s no rigorous public evidence yet on Khanmigo’s effect on learning. Khan Academy’s pre-AI practice platform has research behind it, including a 2026 PNAS study finding modest math gains. The main problem Khan described was low use, not proven ineffectiveness: students can’t benefit from a tutor they don’t use.

Why didn’t students use Khanmigo?

According to Khan Academy officials, many students don’t seek help on their own, and “students aren’t great at asking questions well.” A teacher at an early-adopter school also said students found it frustrating that Khanmigo sometimes made mistakes but wouldn’t give away answers.

What has Khan Academy changed?

Khan Academy rebuilt its platform so Khanmigo appears inside practice problems as a “learning coach” when students get stuck, rather than as a separate chat window. A spokesperson said the change came because students weren’t seeking out Khanmigo’s help as much as hoped. The redesign was set to reach district partners for the 2026 school year.

Should my child use an AI tutor at all?

It can help if it guides rather than answers and if there’s structure around it: a set time, an adult checking in, and a goal. For a child who is significantly behind, prioritize human help, which has much stronger evidence. Teach your child to describe exactly where they’re stuck; that skill makes every tutor more useful.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Barnum, M. (2026, April 9). “Why Sal Khan’s AI revolution hasn’t happened yet, according to Sal Khan.” Chalkbeat. https://www.chalkbeat.org/2026/04/09/sal-khan-reflects-on-ai-in-schools-and-khanmigo/
2. Khan Academy. (2026). “Motivation Meets Mastery: Khan Academy Reimagined for Every Classroom, in Partnership with Districts.” Khan Academy Blog. https://blog.khanacademy.org/khan-academy-reimagined-for-districts-2026/
3. Fesler, L., Martinez Claeys, J. P., Agnew, C., & Loeb, S. (2026). “The Evidence Base on AI in K-12: A 2026 Review.” Stanford SCALE Initiative. https://scale.stanford.edu/sites/default/files/The%20Evidence%20Base%20on%20AI%20in%20K-12%20Report.pdf
4. Eames, T., Brunskill, E., Yamkovenko, B., Weatherholtz, K., & Oreopoulos, P. (2026). “Computer-assisted learning in the real world: How Khan Academy influences student math learning.” PNAS. https://doi.org/10.1073/pnas.2507708123
5. Bastani, H., et al. (2025). “Generative AI without guardrails can harm learning: Evidence from high school mathematics.” PNAS, 122(26). https://doi.org/10.1073/pnas.2422633122
6. Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). “AI tutoring outperforms in-class active learning.” Scientific Reports, 15. https://doi.org/10.1038/s41598-025-97652-6
7. Robinson, C., et al. (2026). “Access Is Not Enough: Human Support Improves Engagement with AI Tutoring.” Working paper, Stanford SCALE Initiative. https://nssa.stanford.edu/studies/access-not-enough-human-support-improves-engagement-ai-tutoring